Add Agent OS governance framework samples (ASI-01, ASI-02, ASI-05)

[imran-siddique]

Add Agent OS Governance Framework — Insecure/Secure Code Samples

Overview

Agent OS is an open-source governance kernel for autonomous AI agents. This PR adds insecure/secure code pairs demonstrating how Agent OS mitigates 3 OWASP Agentic Top 10 risks, with plans to expand to all 9 covered risks.

Samples Added

Risk	Vulnerability	Mitigation
ASI-01 (Agent Hijacking)	Unrestricted goal modification	PolicyEngine validates goals against declarative policy
ASI-02 (Excessive Capabilities)	Unrestricted filesystem/network access	CapabilitySandbox enforces ring-based least-privilege
ASI-05 (Insecure Output)	Raw agent output passed to SQL	Hypervisor OutputValidator sanitizes before downstream

Structure

frameworks/agent-os/
├── README.md                          # Framework overview + architecture
├── Dockerfile                         # Build/run container
├── ASI-01-agent-hijacking/
│   ├── README.md                      # Vulnerability + mitigation description
│   ├── insecure.py                    # Deliberately vulnerable agent
│   └── secure.py                      # Agent OS secured version
├── ASI-02-excessive-capabilities/
│   ├── README.md / insecure.py / secure.py
└── ASI-05-insecure-output/
    ├── README.md / insecure.py / secure.py

Coverage

Agent OS covers 9 of 10 OWASP Agentic Top 10 risks. Full compliance mapping: OWASP-COMPLIANCE.md

Ecosystem

• 82+ GitHub stars, 30+ forks, 5 PyPI packages
• Published on npm (MCP server), listed on Glama
• Active community with external contributors